Evaluating Numeric Rating Scales: Replicated Results

Roberta L. Sangster and Fern Willits


The use of numeric rating scales is very common in survey research. However, studies have found that responses vary by changing numeric scale labels from all positive integers (0 to 10, continuum scale) to include negative and positive integers (-5 to +5, bipolar scale) (Schwarz et al., 1991). The intent of this research is to generalize Schwarz et al's finding across different questions and samples using replicated experiments. The data comes from 8 mail or self-administered surveys that together contain 31 experiments. In general, people are less likely to choose zero or negative integers from a bipolar scale than equidistant integers on a continuum scale. The consistency of the pattern is remarkable across the mail and self-administered surveys. The reluctance to select negative integers can mean that part of a bipolar scale will be virtually unused. In such cases, the spread of the scale values is attenuated, leading to a relatively high mean score and a reduced variance. However, this disappears for less skewed data. We suggest several ideas for designing and using continuum and bipolar measures based on these results.